Bug#835222: needs multiple restarts until network is fully up

[Felipe Sateler]

On 18 October 2016 at 11:13, Marc Haber <mh+debian-bugs at zugschlus.de> wrote:
> Version: 231-9
> thanks
>
> On Tue, Aug 23, 2016 at 02:34:56PM -0300, Felipe Sateler wrote:
>> On 23 August 2016 at 13:22, Marc Haber <mh+debian-bugs at zugschlus.de> wrote:
>> > this is a monster. But relax. Severity: minor.
>> >
>> > I am doing virtualization on an AMD APU machine (pc-engines.ch APU2).
>> > The network setup is rather complex, with multiple physical
>> > interfaces, bridges and VLANs. systemd-networkd is unable to bring up
>> > this configuration in the first try most of the time. About once in
>> > ten tries, it works. In the other cases, multiple invocations of
>> > systemd-networkd restart are needed until the machine has its network
>> > fully initialized.
>>
>> The notworking.txt file has this on all failed starts, and it is not
>> present in the working start nor the working.txt file:
>>
>> Aug 23 16:20:33 prom systemd-networkd[XXX]: Event source
>> 0x55ff6e8ec070 returned error, disabling: No buffer space available
>>
>> So I suppose your config is simply too large and some queue is filling
>> up as systemd-networkd is taking too long to process it.
>> Unfortunately this message is very unhelpful (it was improved in
>> version 229[1]),
>> so it is hard to guess which queue it is.
>
> I was not able to reproduce this at all in a VM, and not on another
> (older) APU with identical software and network setup. I have thus
> done some experiments with the live system and can confirm that using
> a newer systemd (I simply added stretch to sources-list and did apt -t
> stretch install systemd/stretch which surely pulled in some extra
> libraries as well) helps, and the phenomenon is gone.

Thanks for confirming that the problem is not present in newer systemd versions.

>
> After reverting to the LVM snapshots made before these experiments,
> the issue is back again. As I am now reasonably sure that the issue
> will be gone in stretch, I am marking this bug as solved with version
> 231-9, the systemd version currently in stretch.

FWIW, there is a backport of 230-7 in jessie-backports, this might be
useful too, depending on your feelings towards backports.

-- 

Saludos,
Felipe Sateler